AMBER Archive (2007)

Subject: Re: AMBER: Scaling for octahedral versus rectilinear boxes

From: Robert Duke (rduke_at_email.unc.edu)
Date: Fri Aug 31 2007 - 09:23:11 CDT


Hi David -
True. Performance on these systems is never a matter of one quick fix. On
10, I'll do the best I can, but am not yet making external predictions; lots
of time has been spent on exploring performance-enhancement options, but
functionality additions have also eaten up a bit of time. We have also been
impacted by the shift in the industry to unbalanced multicore architectures
with even poorer interprocessor communications characteristics and generally
more problems with shared components (multicore in the short term makes
things worse for us because we get 2/3 of the clockspeed we had and more
memory contention).
Regards - Bob

----- Original Message -----
From: "David Cerutti" <dcerutti_at_mccammon.ucsd.edu>
To: <amber_at_scripps.edu>
Sent: Friday, August 31, 2007 9:27 AM
Subject: Re: AMBER: Scaling for octahedral versus rectilinear boxes

> The reason I ask is that I've noticed that many systems aren't actually
> that much better off in a truncated octahedron versus an orthogonal
> box--for my Streptavidin system, for example, the difference in atom count
> is 20%. In that system, the longest dimension of the molecule is about
> 1.2 and the system is fairly boxy-shaped to begin with. For a system like
> Barnase:Barstar, where the ratio of the longest to the next longest
> dimension is much larger, perhaps as much as 1.6. In the latter case, I
> would expect the orthonormal box to actually solvate the system with fewer
> atoms.
>
> Dave
>
> P.S. Any word on how much better AMBER10 PMEMD will scale than AMBER9?
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber_at_scripps.edu
> To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu
>

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu